Dr. Yu-ke Li is a postdoctoral researcher at California PATH, UC Berkeley. He mainly focuses on the field of computer vision and artificial intelligence, especially in the topic of Machine Learning for Autonomous Driving. Prior to joining PATH, Dr. Li did research on crowds behavior analysis, video understanding and applications on real world requirements, such as autonomous vehicles. He obtained his Ph.D in Wuhan University, China. He also spent his career in France, Italy, Canada and China as a researcher.
Dr Li’s primary research collaborators at PATH are Dr. Ching-Yao Chan, who is the Program Leader at PATH and the Associate Director of Berkeley DeepDrive, and Dr. Pin Wang.
SELECTIVE LIST OF CURRENT AND PAST RESEARCH PROJECTS
Transferring Semantic Knowledge Representation for Urban Scene Understanding
Learning disentangled semantic scene-invariant knowledge representations among different urban scenes
- Disentangling the semantic scene-invariant knowledge from the scene-variant part of the urban scene.
- Integrating the learnt representations to a generative model for enabling semantic knowledge transfer to the urban scene from the target domain.
Dynamic Urban Scene Understanding with Deep Learning
Modeling complex urban motion flows generated from a mixture of traffic(e.g., pedestrians, vehicles, bicycles, wheelchairs)
- A suite of deep reinforcement learning based algorithms capable of distinguishing different forms of dynamic agents (e.g., pedestrians vs cars), and identifying their trajectories within the 3D scene.
- A modeling framework that will automatically derive a generative model from these data that can be used for prediction and planning.
Video forecasting with Generative Adversarial Networks
Foreseeing future video sequence in an unsupervised adversarial training manner
- Harness the spatiotemporal consistency within a video sequence in reciprocal directions for better forecasting.
- Overcome the issues of neglecting dependencies between motion dynamics and visual representations, which existed in two-stream state-of-the-art methods.
Pedestrians behavior understanding in crowds with deep neural networks
Using deep learning approaches to study the motion pattern of pedestrians under crowded scenario
- Provided a well trained feature for motion representation in crowds. Based on this feature, we proposed a simple yet efficient method for path prediction in crowds.
- Explored the spatio-temporal dependencies in crowds with one unified end-to-end deep learning based framework. The proposed framework are tested with two applications: forecasting the future and holistic crowd behavior classifications.
Exploring spatio-temporal interaction among multiple pedestrian tracking
Modeling social interaction by deeply analyzing spatio-temporal context
- Analysis spatiao-temporal context based on evolving spatial information (distance,velocity for several frames)
- Modeled social force based on spatiao-temporal context
- Utilized social force to handled inter-person occlusion for multi-pedestrian tracking
Which Way Are You Going? Imitative Decision Learning for Path Forecasting in Dynamic Scenes
Yu-ke LI , Oral presentation, CVPR 2019 (5.6% acceptance rate) 2019
Video Forecasting with Forward-Backward-Net: Delving Deeper into Spatiotemporal Consistency
Yu-ke LI, Full research paper, 26th ACM Multimedia 2018 (27.5% acceptance rate) 2018
A Deep Joint Spatiotemporal Perspective for Crowd Behavior Understanding
Yu-ke LI, IEEE Transactions on Multimedia 20 (12), 3289-3297, 2018
Pedestrian Path Forecasting in Crowd: A Deep Spatio-Temporal Perspective
Yu-ke LI, Full research paper, 25th ACM Multimedia 2017 (28% acceptance rate), pp. 235-243 2017
Where are you going? An agent Inclusive approach for Path Prediction in Crowd
Yu-ke LI, Wei-ming SHEN, SPIE Journal of Electronic Imaging, 26(4), 043020 2017
Social Interaction based Handling Inter-Person Occlusion for Online Multi-Pedestrian Tracking
Yu-ke LI, Wei-ming SHEN, 12th IEEE International Conference on Advanced Video and Signal-based
Handling Inter Object Occlusion for Multi Object Tracking based on Attraction Force Constraint
Yu-ke LI, Isabelle BLOCH, Wei-ming SHEN, International Conference on Image Analysis and Recognition